Method and device for generating true orthoimage based on three-dimensional gaussian satellite image
By using explicit scene representation and parallel projection rendering based on 3D Gaussian, the problem of high data acquisition cost and low computational efficiency in existing technologies is solved, and the ability to generate true orthophotos with high precision and speed is achieved, which is suitable for large-scale, high-resolution satellite imagery applications.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING UNIV OF CIVIL ENG & ARCHITECTURE
- Filing Date
- 2026-04-08
- Publication Date
- 2026-06-26
AI Technical Summary
Existing technologies suffer from high data acquisition costs, long processing cycles, and high computational costs when generating true satellite imagery, especially in large-scale, high-resolution satellite imagery applications. Traditional methods rely on DSM or LiDAR data, resulting in inaccurate representation of building boundaries, while NeRF-based methods have low training efficiency.
We adopt an explicit scene representation based on 3D Gaussian, reconstruct the 3D scene through a multi-view satellite image set, and generate true orthophotos by using explicit 3D Gaussian voxels and parallel projection rendering, combined with joint optimization and iterative optimization mechanisms. This avoids dependence on expensive elevation data and high computational overhead of volume rendering.
It significantly reduces data acquisition costs and processing cycles, improves the accuracy of building boundary representation, and greatly enhances image rendering efficiency, adapting to the practical application needs of large-scale, high-resolution satellite imagery.
Smart Images

Figure CN122289585A_ABST